DITTO (Bittensor SN118) — Full Analyst Report

SubnetAIQ Intelligence | August 9, 2026 Rating: SPECULATIVE WATCH | Conviction Score: 36/100 | Momentum: SLIGHT NEGATIVE (-26.9) | Verified by independent AI review

EXECUTIVE SUMMARY

Ditto is the memory and identity layer for AI agents on Bittensor (SN118). Built by Omni Aura LLC (team: Seby, Peyton), Ditto provides persistent, cross-agent memory via a bipartite knowledge graph that "dreams" — consolidating fragmented conversation memories into structured knowledge during post-session processing. It integrates with Claude Code, Cursor, Hermes, Codex, and other AI tools via MCP (Model Context Protocol), allowing agents to remember users across sessions, threads, and platforms.

The product is live with 2,300+ users, 55,900+ prompts, iOS/Android apps, and a working MCP server. Mining went live in late June 2026 with an open-sourced agent harness and DittoBench scoring framework. However, on-chain metrics tell a cautionary story: only 6,554 TAO staked (ranked ~50th), conviction score of 36/100 (WATCH), and the alpha token has dropped 47.7% in 30 days from its May highs. The early mover detector flags a REVERSAL pattern (dropped -35%, bounced +24%), but momentum remains negative (-26.9).

We rate SN118 a SPECULATIVE WATCH with a 12-month base-case target of $2.83/alpha (43% upside). The memory-for-agents thesis is compelling, but Ditto faces formidable centralized competitors (Mem0, Zep, Letta) with deeper funding, and the subnet's on-chain traction remains thin.

1. COMPANY OVERVIEW

What Ditto Does

Ditto solves the "Clawzheimer's" problem — AI agents forget everything between sessions. Every time you start a new Claude thread or switch from Cursor to Claude Code, your agent starts from zero. Ditto provides a shared memory layer that follows users across every agent and app they use.

The system works through three layers: 1. Memory Capture: During conversations, Ditto identifies durable facts, preferences, decisions, and corrections worth remembering 2. Knowledge Graph: Memories are stored in a bipartite graph (subjects + memory pairs) in PostgreSQL, with relationships determined by co-occurrence rather than pre-defined schemas 3. Dreaming Pipeline: Post-conversation, an LLM re-reads the session, extracts durable topics, embeds them, searches for duplicates (0.75 cosine similarity merge threshold), and stitches new memories into the graph. A refinement pass synthesizes fragmented descriptions periodically

How the Subnet Works

Miners submit agent-memory harness implementations. Validators score them on tool-calling accuracy and memory recall against fresh randomized datasets. The best harness earns emissions. DittoBench provides a local practice environment that mirrors production validator scoring exactly — "the local practice loop and the on-chain scoring loop are the same loop."

Team & Background

Seby and Peyton are the public-facing team members, appearing in the Revenue Search livestream (May 12, 2026) discussing persistent memory for AI agents. The parent entity is Omni Aura LLC, operating under the principle "Barriers Removed." Limited public information exists about team backgrounds or prior exits.

Note on Escrow Incident: Research did not surface a confirmed "escrow scam" incident specific to Ditto/SN118. There was a separate Bittensor-wide PyPI supply chain attack in 2024 that affected multiple wallets (~$8M stolen), but this was not Ditto-specific. If a Ditto-specific escrow incident occurred, it may have been discussed in Discord/Telegram rather than indexed publicly. This warrants further investigation.

2. TECHNOLOGY DEEP DIVE

Knowledge Graph Architecture

• Bipartite Design: Two node types (subjects and memory pairs) connected by a single junction table in PostgreSQL
• Relationship Discovery: Connections emerge from co-occurrence in memories rather than pre-defined ontologies
• Dreaming Pipeline: Post-session LLM pass extracts topics, embeds them, deduplicates (0.75 cosine threshold), and merges into the graph
• Composite MLP v2 Ranker: Seven retrieval signals, three derived from the knowledge graph (topic prevalence, semantic matching, topic clustering)
• Refinement Pass: Periodic synthesis of fragmented descriptions to improve coherence

Agent Harness (Open-Sourced June 2026)

• Single binary deployment with no external dependencies
• DittoBench starter kit with bundled seed data
• Built-in scoring modes matching on-chain validator logic
• Tunable parameters in a single config file
• Dual-licensed: community open-source + commercial terms

MCP Integration

Ditto's MCP server enables any MCP-compatible AI tool to read/write shared memory: - save_memory: Store durable facts with source context - search_memories / fetch_memories: Retrieve relevant memories before answering - search_subjects / search_memories_in_subjects: Entity-focused lookups - Knowledge graph operations: Create dedicated graphs, publish/subscribe across graphs - Zero LLM cost: MCP users bring their own AI; Ditto handles only memory

Key Technical Differentiators

3. TRACTION & PRODUCT

User Metrics

Product Features

• Workspace Management: Organizations, agents, people, projects, tasks
• File Management: Docs, notes, spreadsheets, images attached to agent threads
• Live Voice Mode: Text-to-voice sessions with maintained context
• Google Workspace Integration: Email, calendar, docs
• Cross-Platform Sync: Phone, tablet, desktop with synchronized context

Revenue Model

Ditto does not appear to have significant direct revenue. The MCP integration is free (zero LLM cost for users). The subnet is funded primarily through TAO emissions. No subscription pricing or enterprise contracts have been publicly announced.

4. ON-CHAIN FINANCIALS

Current Metrics (August 9, 2026)

Price History

SubnetAIQ Scores

Score Breakdown Analysis

The 36/100 conviction score is driven by: - Development: 0/20 — No GitHub activity detected by the scorer (possible detection gap since Ditto's repos are under ditto-assistant org) - On-chain Health: 17/25 — Decent but not exceptional - Market Metrics: 3/15 — Very low liquidity and volume - Valuation: 7/15 — Potentially undervalued relative to emissions - Risk: 11/15 — Some deregistration risk - Importance: 1/10 — Not yet recognized as ecosystem-critical - Price Sustainability: 0/5 — Token in sharp downtrend

5. COMPETITIVE LANDSCAPE

AI Agent Memory Market

The AI agent memory market is exploding in 2026, with multiple well-funded centralized competitors:

Ditto's moat: Only decentralized memory layer in the market. Emission-funded = free for users. MCP integration works across any compatible agent. Knowledge graph "dreaming" is architecturally unique.

Ditto's weakness: Zep outperforms on temporal reasoning (63.8% vs Mem0's 49% on LongMemEval — Ditto's score not publicly benchmarked). Centralized competitors have larger teams, more funding, and enterprise sales motion.

Within Bittensor

No other Bittensor subnet directly competes in agent memory. The closest adjacent play is SN1 Apex (Macrocosmos) for general AI capabilities, but it does not focus on persistent memory.

6. MARKET SIZE

Agent memory is a critical infrastructure layer — every agent needs it, but it's currently underpriced because most solutions are free/freemium to drive adoption. The market will monetize through enterprise contracts as agents move from experiments to production.

7. VALUATION MODEL

Revenue Streams

Three-Scenario Valuation

BEAR CASE — Memory Remains Centralized

• Mem0 and Zep dominate with better benchmarks and enterprise sales
• Ditto remains a niche Bittensor tool with <5,000 users
• Emission share declines as competition for emissions intensifies
• Token continues downtrend toward ATL
• Valuation: $1.5M (below current MCap)
• Alpha price target: 0.005 TAO ($1.02)
• Return: -48% from current

BASE CASE — Ditto Becomes the Bittensor Memory Standard

• Adopted by 10-20 Bittensor subnets as shared memory infrastructure
• User base grows to 20,000+ through MCP virality
• Enterprise tier launches at $29-99/mo
• Emission share stabilizes at 3-5%
• Revenue: $500K-1M ARR
• Valuation: $15M
• Alpha price target: 0.01382 TAO ($2.83)
• Return: +43% from current

BULL CASE — Cross-Ecosystem Agent Memory Standard

• MCP adoption makes Ditto the default memory layer for Claude/Cursor/Codex ecosystem
• 100K+ users, enterprise contracts with AI agent companies
• Partnerships with major agent frameworks (LangChain, CrewAI, AutoGPT)
• Revenue: $5-10M ARR, 15x multiple
• Valuation: $75-150M
• Alpha price target: 0.069 TAO ($14.14)
• Return: +617% from current

Probability-Weighted Expected Value

8. PRICE TARGETS

Assumes constant TAO price at $204.91. TAO appreciation amplifies USD returns.

9. RISK FACTORS

CRITICAL RISKS

Token in Severe Downtrend (CRITICAL) - Price has dropped -70.2% from ATH in less than 3 months - 30-day return of -47.72% signals sustained selling pressure - Price sustainability score of 0/5 — worst possible rating - Momentum score of -26.9 with -81.5 price momentum

Centralized Competition with Better Benchmarks (CRITICAL) - Zep scores 63.8% on LongMemEval vs Mem0's 49% - Ditto has no published benchmark results - Centralized competitors have larger teams, more funding, and enterprise sales teams - Memory is a commodity layer — winner-take-most dynamics likely

HIGH RISKS

Minimal On-Chain Traction (HIGH) - Only 6,554 TAO staked (ranked ~50th out of 128 subnets) - Conviction score 36/100 is one of the lowest in the portfolio universe - Development score of 0/20 suggests limited visible activity

Revenue Model Unclear (HIGH) - No clear monetization beyond TAO emissions - MCP integration is free — how does Ditto capture value? - Enterprise tier not yet announced - $0 revenue against $384K/year in emission subsidies

Liquidity Risk (HIGH) - Extremely thin trading volume - Market cap of ~$4M with low liquidity pool - Any meaningful position faces severe slippage

MEDIUM RISKS

Team Transparency - Small team (Seby, Peyton) with limited public background - No known VC backing or institutional investors - Omni Aura LLC is a young entity

Bittensor Emission Dependency - 33,917% emission APY means massive supply inflation - If emissions decline or get redirected, the token faces structural sell pressure - Staker APY of 30.9% is healthy but doesn't offset the -47.72% monthly price decline

Unresolved Escrow Incident - User reports of an escrow-related issue exist but details remain unconfirmed - Could not verify through public sources — warrants direct community investigation

10. ANALYST OPINION

Rating: SPECULATIVE WATCH

Why WATCH (not BUY): 1. Token in freefall — down 70% from ATH with SLIGHT NEGATIVE momentum 2. Conviction score 36/100 — well below the 60+ threshold for actionable positions 3. Zero development score — either no activity or detection gap 4. No revenue — emission-funded with no clear path to monetization 5. Better alternatives exist — centralized memory solutions (Zep, Mem0) have stronger benchmarks and traction

Why not SELL / AVOID: 1. Thesis is compelling — agent memory is a real, growing market ($300-500M in 2026) 2. MCP moat — only decentralized option; MCP virality could drive organic adoption 3. Dreaming pipeline is architecturally novel — no competitor does post-session knowledge graph consolidation 4. Early mover REVERSAL pattern — dropped -35%, bounced +24%. Could be forming a bottom 5. Volume momentum +68.4 — someone is accumulating despite price decline

Entry Strategy: - Wait for conviction score to improve above 50 before initiating a position - Watch for: user growth acceleration, benchmark publication, enterprise tier launch, emission share increase - If entering speculatively: maximum 1-2% of Bittensor portfolio allocation - Stop-loss: Below ATL of $0.52 (0.002546 TAO)

Key Catalysts to Watch: 1. DittoBench v9 is live — benchmark results vs Mem0/Zep are the single most important catalyst 2. heyditto.ai has crossed 10,000+ conversations and shipped public file sharing — watch for next user growth milestone 3. Enterprise tier pricing announcement 4. Adoption by other Bittensor subnets as shared memory infrastructure 5. User growth from 2,300 toward 20,000+ 6. GitHub activity spike (would fix the 0/20 development score)

APPENDIX: DATA SOURCES

• On-chain data: Bittensor metagraph, SubnetAIQ live data, TaoStats OHLC cache
• Product data: heyditto.ai website, Ditto MCP server tools
• Technical architecture: TAO Daily articles on dreaming pipeline and mining stack
• Team data: Revenue Search livestream (May 12, 2026)
• Competitor benchmarks: Mem0 blog (State of AI Agent Memory 2026), Medium comparisons
• Market sizing: Industry reports, agent market forecasts
• SubnetAIQ engines: Conviction Scorer v2.2, Momentum Engine, Early Mover Detector

This report is for informational purposes only and does not constitute financial or investment advice. Cryptocurrency investments carry significant risk. Past performance does not guarantee future results.

SubnetAIQ — First to know. First to move.